JP2008263177A - Software sequencer for integrated substrate processing system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an apparatus and a method for scheduling a process sequence to achieve maximum throughput and process consistency in a cluster tool having a set of constraints. <P>SOLUTION: The method for scheduling a process sequence comprises a step for determining an initial individual schedule by assigning resources to perform the process sequence, a step for calculating a fundamental period, a step for detecting resource conflicts in a schedule generated from the individual schedule and the fundamental period, and a step for adjusting the individual schedule to remove the resource conflicts. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

Background of the Invention

Field of Invention
[0001] Embodiments of the present invention generally relate to an apparatus and method for transferring a semiconductor substrate in an integrated processing system. More specifically, embodiments of the present invention relate to an integrated substrate processing system having a software sequencer that provides inter-substrate timing consistency.

Explanation of related technology
[0002] In current semiconductor processing, multilayer members are fabricated on a semiconductor substrate using a specific processing recipe having multiple processing steps. Cluster tools integrate a number of process chambers for performing a process sequence without removing the substrate from the processing environment, usually a controlled environment, and are generally used in processing semiconductor substrates. A process sequence is generally defined as a sequence of device fabrication steps or process recipe steps that are completed in one or more processing chambers of a cluster tool. A process sequence may generally include electronic device fabrication processing steps for various substrates (ie, wafers).

  [0003] Over the years, semiconductor device manufacturers have faced a dilemma between system throughput and process consistency to remain competitive.

  [0004] On the other hand, the effectiveness of the substrate fabrication process directly affects the competitiveness of device manufacturers. On the other hand, the reduction in semiconductor device component size has resulted in semiconductor manufacturing specifications with more stringent requirements for process consistency and repeatability.

  [0005] The effectiveness of the substrate fabrication process is often measured by two related important factors: device yield and cost of ownership (CoO). CoO is affected by a number of factors, but is greatly influenced by system and chamber throughput, or simply by the number of substrates processed per hour using the desired processing sequence.

  [0006] In an attempt to reduce CoO, electronic device manufacturers often seek to optimize process sequences and chamber processing times to achieve the maximum substrate throughput possible in view of cluster tool architecture limitations and chamber processing times. Spend time. System throughput may be increased by reducing chamber limits and / or robot limits. The chamber limit is determined by the time consumed by the longest process recipe step in the processing sequence. The robot limit is determined by the time spent for transferring the substrate by the robot.

  [0007] For some processing sequences, such as heat treatment and wet processing, process consistency and repeatability are closely related to timing consistency. Time consistency can be achieved by good control over the substrate transfer time between chambers and the substrate cue time, which is the time the substrate remains in the chamber after the recipe step.

  [0008] Semiconductor manufacturing may require a trade-off between throughput and process consistency. For example, on the one hand, adding substrate cue time within cue time sensitivity limits between processing steps is an effective way to achieve process consistency and repeatability. On the other hand, the addition of substrate cue time may increase chamber / robot limitations resulting in reduced system throughput.

  [0009] In order to ensure uniform substrate surface properties, it is desirable that all substrates have consistent timing at each step of the process sequence. In modern systems, limited precedence algorithms are used in board scheduling to prevent deadlocks. A limited predecessor algorithm can stabilize the system to a maximum throughput value. After reaching steady state, all substrates have the same cue time at each step. However, until reaching steady state, different substrates will behave differently based on modern systems. For example, the first substrate has no latency because all resources are free at this point. However, subsequent substrates must wait in some steps. In addition, the substrate cue time is determined by the steady state and there is no way to constrain the cue time. Thus, some processing steps with high cue time sensitivity may be compromised in steady state.

  [0010] Therefore, there is a need for a cluster tool for an apparatus and method for determining optimal throughput and process consistency.

Summary of the Invention

  [0011] Embodiments of the present invention generally provide an apparatus and method for scheduling process sequences to achieve maximum throughput and process consistency in a cluster tool with a set of constraints.

  [0012] One embodiment of the present invention is the step of determining an individual schedule by allocating resources to execute a process sequence, wherein the individual schedule assigns each of a plurality of process steps of the process sequence to an individual substrate. A start time at which the base period is calculated, a basic period is calculated, wherein the basic period is defined as the time between the start times of two successive substrates, the individual schedule and A method is provided for scheduling a process sequence comprising: detecting a resource collision in a schedule generated from the fundamental period; and adjusting the individual schedule to remove the detected resource collision.

  [0013] Another embodiment of the present invention, when executed by a process, is the step of determining an individual schedule by allocating resources to execute the process sequence, wherein the individual schedule is included in the process sequence. Each of a plurality of process steps includes a step having a start time at which an individual substrate starts, and a step of calculating a basic period, wherein the basic period is defined as a time portion between two continuous substrate start times. Performing operations comprising: detecting a resource collision of the schedule generated from the individual schedule and the base period; and adjusting the individual schedule to remove the detected resource collision A computer program that schedules the process sequence A computer-readable medium containing a gram.

  [0014] Yet another embodiment of the present invention is a step of generating a processing schedule, the step having no waiting time for each of the plurality of processing steps of the processing sequence, and the basic period according to the usage period of the bottleneck resource. Determining, detecting a resource collision of the processing schedule based on the basic period, and adjusting at least one of the processing schedule and the basic period to remove the detected resource collision A processing sequence scheduling method is provided.

  [0015] In order that the above-cited features of the present invention may be understood in detail, a more specific description of the invention briefly summarized above may be made by reference to embodiments, some of which are It is illustrated in the accompanying drawings. The accompanying drawings, however, illustrate only typical embodiments of the invention and, therefore, the invention may recognize other equally effective embodiments and should not be deemed to limit this scope. It should be noted.

Detailed Description of the Preferred Embodiment

  [0025] Embodiments of the present invention generally provide an apparatus and method for processing a substrate using a multi-chamber processing system. More specifically, embodiments of the present invention provide a method for scheduling a given process sequence. The scheduling method of the present invention allows all substrates in the system to ensure uniform substrate characteristics with a consistent cue time at each step of the process sequence. The scheduling method of the present invention includes a step of determining a schedule by allocating resources to a cluster tool to execute a given process sequence, and a basic period and a cluster tool according to the length of the bottleneck process step and the transfer motion. Determining the time between sending two successive substrates. The method further includes checking resource collisions of the determined schedule using the determined fundamental period; adding queue time to the schedule and / or extending the fundamental period to eliminate resource collisions With.

  [0026] Resource collisions are detected and eliminated by reducing periodic system problems and solving a set of equations within a (0, T) time interval, where T denotes a predetermined fundamental period. Yes. In one embodiment, a game tree algorithm is used to resolve resource conflicts. In one embodiment, an efficient method of game tree timing is used to find the first viable solution.

  [0027] Embodiments of the present invention are described herein according to a polysilicon generation sequence. FIG. 1 schematically illustrates a semiconductor processing cluster tool 100 according to an embodiment of the present invention. It is envisioned that the method described herein may be practiced with other tools that are configured to perform a process sequence.

  [0028] The cluster tool 100 includes a vacuum-tight processing platform 101 and a factory interface 102. Platform 101 includes a plurality of processing chambers 110, 108, 114, 112, 118, 116 and at least one load lock chamber 120, which are coupled to vacuum substrate transfer chambers 103, 104. The factory interface 102 is coupled to the transfer chamber 104 by a load lock chamber 120.

  [0029] In one embodiment, the factory interface 102 comprises at least one docking station, at least one substrate transfer robot 138, and at least one substrate aligner 140. The docking station is configured to receive one or more front opening unified pods 128 (FOUP). Two FOUPs 128A, 128B are shown in the embodiment of FIG. The substrate transfer robot 138 is configured to transfer a substrate from the factory interface 102 to the load lock chamber 120.

  [0030] The load lock chamber 120 has a first port coupled to the factory interface 102 and a second port coupled to the first transfer chamber 104. The load lock chamber 120 pumps down the chamber 120 as required to facilitate the passage of the substrate between the vacuum environment of the transfer chamber 104 and the substantially ambient (eg, atmospheric) environment of the factory interface 102. Combined with pressure control system to ventilate and.

  [0031] The first transfer chamber 104 and the second transfer chamber 103 have a first robot 107 and a second robot 105 disposed therein, respectively. Two substrate transfer platforms 106A, 106B are located in the transfer chamber 104 to facilitate transfer of substrates between the robots 105, 107. The platforms 106A, 106B may be open relative to the transfer chambers 103, 104, or may be selectively isolated (ie, sealed) from the transfer chambers 103, 104, with different operating pressures being applied to each of the transfer chambers 103, 104. To be maintained at.

  [0032] A robot 107 located in the first transfer chamber 104 can transfer substrates between the load lock chamber 120, the processing chambers 116, 118, and the substrate transfer platforms 106A, 106B. A robot 105 located in the second transfer chamber 103 can transfer substrates between the substrate transfer platforms 106A, 106B and the processing chambers 112, 114, 110, 108.

  [0033] FIG. 2 illustrates a flowchart of one embodiment of a process sequence 200 for depositing a dielectric layer on a substrate in an integrated cluster tool, such as the cluster tool 100 described above.

  [0034] The process sequence 200 begins with step 202 of positioning a substrate on a cluster tool.

  [0035] In step 204, a dielectric layer is deposited on the substrate. The dielectric layer may be a metal oxide and may be deposited by an ALD process, MOCVD process, conventional CVD process or PVD process.

[0036] Following the deposition process, the substrate may be exposed to a post-deposition annealing (PDA) process at step 205. The PDA process may be performed in a high speed annealing chamber, such as a RADIANCE ^™ RTP chamber.

  [0037] In step 206, the dielectric layer is exposed to an inert plasma process to increase the density of the dielectric material to form a plasma treatment layer. The inert plasma process may include a decoupled inert gas plasma process performed by flowing an inert gas through a decoupled plasma nitridation (DPN) chamber.

  [0038] In step 208, the plasma treatment layer deposited on the substrate is exposed to a thermal annealing process.

  [0039] In step 210, a gate electrode layer is deposited on the annealed dielectric layer. The gate electrode layer may be polycrystalline Si, amorphous Si or other suitable material deposited using an LPCVD chamber.

  [0040] Table 1 illustrates the recipe time and chamber requirements for each step of the sequence 200.

  [0041] The method of the present invention relates to determining a process schedule that achieves substrate consistency, stays within resource constraints, and maximizes throughput.

  [0042] The process schedule of the present invention may include a schedule of individual substrates (hereinafter, individual schedule) and a basic period between successive substrates. The individual schedule includes the start time and end time of each process step of the substrate relative to the start time of the initial movement of the substrate. The fundamental period defines the rate at which substrates are sent to the cluster tool. Specifically, the basic period is a time interval between two continuous substrates.

  [0043] Factors affecting the process schedule include the process sequence to be performed, the time it takes to execute each recipe step, the substrate queue time constraint at each step, and the transfer time between different chambers. There is. The substrate queue time constraint is usually part of a sequence that defines the maximum time that a substrate can wait for a given process step in a given chamber after the process recipe is complete. Table 2 illustrates exemplary substrate cue time constraints for associated chambers that can be used to perform the process sequence 200. The time taken to perform each recipe step generally includes timing information regarding the cleaning and periodic cleaning processes. For a simple time-based recipe, the time taken to execute each recipe step may be calculated by analyzing the recipe step. For endpoint-based recipes, statistical information such as the average time may be used for scheduling. The transfer time is the actual robot movement time and overhead for other recipes that are executed as part of the transfer itself.

  [0044] Referring to Table 2, the total recipe time range, in the case of the process sequence 200, indicates the time range for processing the substrate in the corresponding chamber according to a given process sequence. The cue time sensitivity indicates the maximum waiting period that the substrate has in the corresponding chamber after the process step is completed in the chamber. The substrate handling variation limit indicates the maximum variation in cue time between substrates to obtain the desired process consistency. The cleaning frequency indicates the frequency with which the corresponding chamber needs cleaning. The cleaning time relates to the time required to complete the cleaning process. For chambers that require periodic cleaning, the cleaning process may be handled according to the frequency and length of cleaning requirements. For chambers that require cleaning after each substrate, such as DPN + (A) and DPN + (B), the cleaning time is generally added to the process time. Further cleaning considerations are illustrated in FIG.

[0045] In one embodiment of the present invention, the process schedule of the present invention comprises the following steps: allocating resources to perform a given process sequence to determine an initial individual schedule; It may be determined by determining, checking for an initial individual schedule and an initial fundamental period resource collision, and removing a resource collision by adding a substrate queue time to the individual schedule. In one embodiment of the invention, determining the process schedule may comprise extending an initial fundamental period to eliminate resource conflicts.
Resource allocation and initial individual schedule determination
[0046] Assigning resources generally comprises setting up a cluster arrangement of cluster tools and assigning a robot to transfer substrates between the arranged chambers.

  [0047] The chamber arrangement may include defining a chamber position and number of chambers for the process steps. The chamber arrangement may be affected by the process sequence that is performed, the time it takes to perform each recipe step, and the substrate cue time constraint at each step.

  [0048] For example, the cluster tool 100 may be configured to execute the process sequence 200. Appropriate chambers may be selected for the chambers 108, 110, 112, 114, 116, 118 to facilitate the process sequence 200. For example, the chambers 116, 118 may be chemical vapor deposition (CVD) chambers configured to deposit polycrystalline silicon (POLY). One suitable chamber is a POLYGen chamber available from Applied Materials, Inc. Chambers 108, 114 may be decoupled plasma nitridation (DPN) chambers. Chambers 110, 112 may be rapid thermal process (RTP) chambers. One or more cooling chambers may be positioned above the substrate transfer platforms 106A, 106B.

  [0049] In determining the arrangement of chambers in the cluster tool 100, resources including chambers, load locks and robots may be allocated for each process step and transitions between steps.

  [0050] FIG. 3A schematically illustrates a flowchart of an exemplary process sequence in accordance with one embodiment of the present invention. FIG. 3B schematically illustrates the route of the substrate processed in the process sequence of FIG. 3A in the cluster tool 100 of FIG. Referring to FIG. 3A, steps S1-S13 represent a substrate that remains in the process chamber, transfer chamber, or load lock. The movements m1 to m12 represent the movement of the substrate between the chambers conveyed by the robot. Movements m1-m12 are further illustrated by the arrows in FIG. 3B.

  [0051] Table 3 illustrates an individual schedule of the process sequence 200. The process time indicates the total time that the substrate occupies the resource, chamber or robot. The start marks the time when a substrate that occupies resources for the substrate first enters the cluster tool. The end marks the time when the substrate that releases resources to the substrate enters the cluster tool. In the initial individual schedule, no queue time is added to any step. If the substrate is not in the cluster tool, the substrate can follow this schedule. As shown in Table 3, it takes 1233 seconds for the substrate to complete the process sequence 200. If only one substrate is in the cluster tool, a maximum of two resources are occupied at a given time while the remaining resources are idle. To reduce idle time and increase throughput, a second substrate may be provided to the cluster tool before the first substrate exits the cluster tool. The amount of time between sending two substrates, i.e. the fundamental period, may be minimized to maximize throughput.

Judging the initial basic period
[0052] In one embodiment of the present invention, the initial fundamental period may be determined according to the longest usage period between all resources in the cluster tool. The resource usage period may be defined by the total time taken to perform all steps / movements of the process sequence on the signal board.

  [0053] In one embodiment, the duration of use of each resource repeats all process steps of the process sequence and divides each process step into subparts including load time, unload time, process recipe time and cleaning time. May be calculated. Each of the subparts is then assigned to one resource (or multiple resources) required for the subpart.

  [0054] For chambers, if the chamber is required for all steps used in the process sequence, the duration of use may include load time, process recipe time, unload time and cleaning time. If at least two chambers are arranged to perform one step, the period of use may be divided by the number of chambers. In one embodiment, the duration of use of one chamber can be calculated using the following formula:

Here, D [i] indicates a usage period of the chamber i, k indicates a process step in which the chamber i is used, P [k] indicates a process time of the step k, and L [ k] indicates the load time of step k, U [k] indicates the unload time of step k, C [k] indicates the cleaning time of step k, and n is the number of chambers i. Is shown. The sum is for all steps performed in chamber i.

  [0055] For robots, the duration of use may include pick-up time, transfer time and drop time for all movements in which the robot is used. In one embodiment, the duration of use of one robot may be calculated using the following formula:

Here, D [j] indicates a usage period of the robot j, l indicates a movement in which the robot j is used, Pk [l] indicates a pickup time of the movement l, and Tr [l ] Indicates the transfer time of the movement l, and Dr [l] indicates the drop time of the movement l. The sum is for all movements performed by robot j.

  [0056] In one embodiment, the initial fundamental period may be set to the maximum usage period of all resources including chambers and robots.

  [0057] Generally, the pick-up time for robot movement overlaps with the unload time of the preceding step, and the drop time for robot movement overlaps with the loading time of the subsequent step. Thus, the chamber step load time overlaps the drop time of the preceding movement, and the chamber step unload time overlaps the pickup time of the subsequent movement. In order to simplify the chamber duration calculation, the chamber usage period may include the time required for the preceding movement, the time required for the subsequent movement, the process time, and the cleaning time if necessary. Table 4 lists the calculation of the resource usage period of the cluster tool 100 for executing the process sequence 200. As shown in Table 4, the maximum usage period is 240 seconds, which belongs to the RTO chamber. Accordingly, the initial fundamental period may be set to 240 seconds according to an embodiment of the present invention.

Check for resource conflicts
[0058] A resource collision is a situation where one resource is needed by two or more steps or moves simultaneously. Resource collisions may be generated when two or more substrates are in a cluster tool and when one or more resources are used in two or more steps or movements. In general, robot collisions are common because robots are typically used for multiple movements of the process schedule. However, resource collisions can occur in the process chamber, load lock and / or transfer chamber when resources are scheduled in more than one step of the process sequence.

  [0059] In one embodiment of the present invention, resource collisions corresponding to a given individual schedule and basic period are checked by calculating a relative start time and a relative end time for each step / movement in one period. May be.

[0060] In one embodiment, the relative start time S _Relative [i, N] and the relative end time E _Relative [i, N] of step i of the Nth substrate are:

May be calculated. Where i indicates the number of steps / movements, N indicates the substrate sequence number, FP indicates the basic period, and S [i, N] is the absolute start of step i on the Nth substrate. E [i, N] indicates the absolute end time of step i of the Nth substrate. S [i, N] and E [i, N] may be calculated by the following formula:

  Here, i represents the number of steps / movements, N represents the substrate sequence number, FP represents the basic period, and D [j] represents the usage period of the i-th step / movement. .

[0061] In one embodiment, resource collisions may be detected by detecting overlapping intervals of relative start times and relative end times of different steps / movements. For example, if steps i and k require the same resource, the intervals (S _Relative [i, N], E _Relative [i, N]) and (S _Relative [k, N], E _Relative [k, N] ) Indicates a resource conflict.

  [0062] Table 5 lists exemplary resource collision results for the initial individual schedule of Table 4 during a basic period of 240 seconds. As shown in Table 5, M9 of the rear robot collides with M5 and M6, and M12 of the FI robot collides with M1 and M2.

  [0063] FIG. 4 schematically illustrates a recipe diagram for the schedule table of Table 5. As shown in FIG. 4, six substrates are processed in the system. Each substrate is sent separately to the system at a fundamental period. M9 of the first substrate and M6 of the third substrate require a rear robot when the collision 1 is caused, and M9 of the first substrate and M5 of the fifth substrate are the rear robots when the collision 2 is caused. The required M12 of the first substrate and M1 of the fifth substrate require the FI robot when the collision 3 is caused. The first substrate M12 and the fifth substrate M2 require an FI robot when incurring a collision 4.

Remove resource conflicts
[0064] In one embodiment of the invention, resource collisions can be eliminated by adding a queue time to delay one of the two steps involved in the collision. In one embodiment, the cue time may be added to delay the later step of the two collision steps.

  [0065] As shown in Table 6, the rear robot collision between M9 and M5 and between M9 and M6 is eliminated by adding a cue time of 25 seconds to step S9. M9 is delayed by 25 seconds and each board is scheduled to stay in the system for 1258 seconds compared to 1233 seconds before the cue time. However, since the fundamental period remains 240 seconds, system throughput is not reduced due to delay.

  [0066] FIG. 5 schematically illustrates a recipe diagram for the updated schedule table listed in Table 6.

  [0067] In some cases, a new resource collision may be created from a queue. As shown in FIG. 5, a new collision between M10 and M3 and between M12 and M2 is generated as a result of the added cue time. In one embodiment of the invention, an updated schedule table may be generated, resource conflicts may be checked against the updated schedule table, and additional queue time may be added to the new conflict after adding queue time. May be introduced. In one embodiment, the queue time may be added to the individual schedule until there are no resource conflicts. However, in some cases, resource collisions may not be removed by adding queue time, or the added queue time is outside the resource's queue time sensitivity constraints (such as the constraints shown in Table 2). Sometimes. If resource collisions cannot be eliminated by adding queue time to the processing step, the base period may be extended, and resource collisions are checked for the initial individual schedule based on the extended base period. And may be removed.

  [0068] FIG. 6 illustrates a flowchart of a scheduling method 400 according to an embodiment of the invention. The scheduling method 400 is configured to find a schedule for a process sequence. The schedule ensures consistent maximum throughput between substrates and observes resource constraints such as queue time sensitivity. The schedule obtained for the scheduling method 400 comprises an individual schedule and a basic period, the individual schedule shows the timetable of the individual substrates of the cluster tool throughout the process sequence, and the basic period is the start of two consecutive substrates. Shows the time interval between hours. An exemplary schedule is shown in Table 3.

  [0069] In step 410 of the scheduling method 400, an initial individual schedule may be determined for the process sequence. The initial individual schedule comprises a timetable for the substrate in a cluster tool that has no waiting time in any step and movement. The initial individual schedule is generally dependent on the process sequence and the topology of the cluster tool on which the substrate is processed. To be judged.

  [0070] In step 420, an initial fundamental period may be determined. In one embodiment, the initial basic period is set as a usage period of a bottleneck resource such as a chamber or a robot. The embodiment for calculating the usage period is as described above. Setting the initial basic period as the usage period of the bottleneck resource ensures that a possible schedule start is searched from the highest throughput.

  [0071] In step 430, a schedule table may be generated based on the initial individual schedule and the initial fundamental period. In one embodiment, the schedule table may include a time table within a basic period for each resource. For example, in the schedule table related to Table 4, within each basic period (0,240), the FI robot performs M1 at (0,22), M2 at (27,49), and (15,33) It is necessary to execute M12.

  [0072] At step 430, resource conflicts are checked against the generated schedule table. In one embodiment, resource collisions may be determined by checking for duplicate timetables for each resource within the fundamental period. For example, the FI robot time table in Table 4 has an overlap of M1 / M12 and M2 / M12. In one embodiment, resource collision checks may be performed on resources required by at least two steps and / or movements during the process sequence.

  [0073] If a resource conflict is found in the schedule table for all resources in the cluster tool, the individual schedule and base period for the schedule table are acceptable solutions to this problem, and the method moves to step 470 and the process Output the current individual schedule and basic period.

  [0074] If there is a resource conflict in the schedule table, the resource conflict may be removed by adjusting the individual schedule in step 450. In one embodiment, resource conflicts may be eliminated by adding queue time to the individual schedule. In one embodiment, the queue time may be added to delay one of the steps leading to resource collisions. In one embodiment, a game tree algorithm may be used to remove resource conflicts. In one embodiment, resource constraints are considered when adding queue time to an individual schedule. A detailed method for removing resource conflicts is described according to FIG.

  [0075] Step 450 outputs the result. In step 460, the output from step 450 is verified. If the resource conflict is eliminated by adjusting the individual schedule, the scheduling method moves to step 470 and outputs the updated individual schedule and the current fundamental period. However, if the resource conflict cannot be removed by adjusting the individual schedule, the scheduling method proceeds to step 480.

  [0076] In step 480, the current fundamental period is extended. In one embodiment, the fundamental period may be extended by a predetermined increment. By extending the fundamental period, the scheduling method searches for possible solutions in areas where throughput is reduced.

  [0077] In step 490, an updated schedule table is generated from the extended basic period and the initial individual schedule, in which case no queue time is added. The scheduling method then proceeds to step 440 to check for resource conflicts.

  [0078] Accordingly, the scheduling method 400 provides a schedule with inter-substrate consistency and maximum throughput for a given process sequence.

Game tree algorithm
[0079] In one embodiment of the present invention, the game tree algorithm may be used in a scheduler, such as scheduling method 400, to eliminate schedule table collisions.

  [0080] The game tree concept is used in the game theory to determine the most likely move that the system can win a given game. The game tree is a directed aperiodic graph, where each node of the aperiodic graph is the state of the system, eg, individual schedule and basic period, and each edge is moved, eg, changed to an individual schedule or basic period. Represents. A pair of nodes and the edge connecting the pair of nodes can be considered a differential change that occurs in the system when a move is performed. At each step of the game, the most likely move is selected by searching the game tree.

[0081] A game tree may be viewed as a data structure used to solve an associative problem. In this particular case of scheduling, resource collisions such as robots can be resolved by rearranging the start and end times of one step when two or more steps overlap. The next move that the system should head for resolution is selected with the best reordering order to minimize resource collisions. However, since this algorithm is in factorial order, the required computational resources are very large when multiple movements are considered. Therefore, an efficient way to reduce the number of possibilities to consider is necessary to make this solution feasible. Reducing the number of possibilities to consider is referred to as game tree trimming. In one embodiment of the invention, tree trimming may be accomplished by using a maximum queue time constraint to eliminate multiple possibilities. The maximum queue time constraint may be a predetermined time and may be given by the target process sequence. The maximum queue time may also be selected by the scheduler based on the fundamental period if not predetermined. For a given resource, the maximum queue time may be selected using the following formula:
Maximum queue time = maximum value of (bottleneck resource usage period-(target resource usage period), sequence user-defined value)
[0082] In one embodiment of the invention, the game tree is used to find a first viable solution. By choosing the first viable solution instead of the best solution, this problem is greatly simplified. The purpose of the scheduler is to maximize throughput. When the basic period is set, the queue time is constrained by the above maximum queue time formula, so the added queue time does not change the bottleneck period. Cannot be changed. Thus, the problem of resolving resource conflicts counters maximizing throughput. The best solution would be a solution that, if found, minimizes the queue time required at each step. However, this only leads to a negligible gain over long production batches. For efficiency, if a first viable solution is found, determine the end of the search.

  [0083] With the above simplification, the algorithm need not build a complete search tree. The method of the present invention starts building the search tree and rejects some branches that violate the maximum queue time constraint if the branch leads to cyclic dependencies, ie, resource collisions reoccurring. If the first branch of the search tree that resolves all conflicts is found, the algorithm ends and uses this as the solution.

  [0084] A game tree according to the present invention may be created with depth in mind. The scheduler iterates through all process steps / movements and allocates resources for each step / movement execution. If multiple resources are available for step execution, the scheduler multiplexes the available resources to achieve a uniform load distribution. After allocating resources, the scheduler checks for resource collisions. To do. To identify recurring resource conflicts, the scheduler also maintains a history of all resource conflicts that it has resolved. Before delaying any step, the scheduler queries the collision history table to see if the same resource collision has been resolved in the past. If resource conflicts have been resolved in the past, the scheduler rejects the change and tries to find another viable solution. To trim the game tree, the calculated delay is compared to the maximum cue time constraint for a given step. This step is only delayed if the constraints are not violated.

  [0085] In order to resolve the resource conflicts of steps I and K, the scheduler has two ways to eliminate this resource conflict: step I delay and step K delay. These two solutions lead to two different branches of the game tree. In one embodiment of the invention, the scheduler first attempts to delay the step with the earlier relative start time. If step K has a relative start time earlier than step I, the scheduler attempts to add a queue time to delay step K first. If the collision history table does not have the same collision and the maximum queue time constraint is met, a new child node is created and the schedule table is adjusted to reflect the new delay in step K . If step K is delayed by Δ seconds, the start times of all steps following step K, which are step K + 1 to step N, where N is the total number of steps, are also increased by the same Δ.

  [0086] After resolving the resource conflict, the scheduler searches for a resource conflict in the updated schedule table after the queue time has been added. Resource conflicts may be created because of delays introduced to eliminate resource conflicts, or may already exist.

  [0087] If step K cannot be delayed due to any constraint violation, the scheduler moves to the next branch by attempting the delay of step I. This is sometimes referred to as collision reversal. If the possibility of both Step K delay and Step I delay is rejected, the scheduler moves the game tree up to reverse the parent node collision. Since each resource conflict can be resolved in exactly two ways, a count of the number of attempts to resolve the conflict may be used. If the count exceeds 2, the current node has no solution and the scheduler moves the game tree up and tries and reverses other collisions in the branch.

  [0088] If resources are allocated to all steps and there are no conflicts, the solution is accepted and the delay of each step is used by the scheduler.

  [0089] In some cases, there is no set of resource collision solutions. The game tree method does not send back a solution, and the scheduler then increases the fundamental period by a small difference and recreates the relative time in the scheduler table that uses the extended fundamental period.

  [0090] FIG. 7 illustrates a flowchart of a resource collision removal method 500 according to an embodiment of the present invention. The method 500 is configured to search for a first viable solution and eliminate resource conflicts for a given process sequence and a given base period. The game tree theory described above is used in method 500. Method 500 may be used in step 450 of scheduling method 400 of FIG.

  [0091] In step 502, a schedule based on the individual schedule and the fundamental period is provided. The schedule has resource conflicts. In one embodiment, resource conflicts may be detected from the schedule.

  [0092] In step 504, the current collision is the resource collision to be removed and is set to the first resource collision in the schedule. In one embodiment, the first resource collision is defined by a first collision contact in the timetable of the individual schedule. For example, collision 1 in FIG.

  [0093] In step 506, the collision history may be searched to check whether the current collision has already been resolved.

  [0094] If the current collision is not in the collision history, the current collision is added to the collision history at step 508.

  [0095] In step 510, a first solution for removing the current collision is attempted. Step 510 may include calculating the cue time required to delay the earlier start time step to remove the current collision. In one embodiment, the counter for the current collision is set to 1 to mark the number of attempts made to resolve the current collision.

  [0096] In step 512, the calculated queue time may be compared to a maximum queue time constraint. In one embodiment, a resource maximum queue time constraint may be defined by a minimum value of a user defined constraint and a difference between a bottleneck resource usage period and a resource usage period.

  [0097] If the calculated queue time is within the maximum queue time constraint, the current individual schedule is calculated to delay the step with the earlier start time of the current collision as shown in step 514. It may be updated by adding a queue time.

  [0098] In step 516, the collision may be checked for the updated individual schedule and the current fundamental period. If there is no collision, method 500 has found a solution. The current individual schedule and current basic period may be output as a solution at step 520. However, if a collision is detected at step 516, the method proceeds to step 518 where the current collision is set to the first collision of the updated schedule. The method then moves back to step 506.

  [0099] Referring back to step 512, if the calculated queue time does not meet the maximum queue time constraint requirement, the algorithm moves to step 522.

  [0100] In step 522, the current collision is reversed, in which case the cue time required to delay the step with the slower start time and remove the current collision is calculated. In one embodiment, a counter for the current collision is set to 2 to mark the number of attempts made to resolve the current collision.

  [0101] In step 524, the queue time calculated from step 522 is compared to a maximum queue time constraint. If the calculated queue time satisfies the maximum queue time constraint, the method moves to step 514 and updates the current individual schedule. However, if the calculated cue time does not satisfy the maximum cue time constraint, the current collision is removed from the collision history at step 526.

  [0102] In step 528, the collision history is checked. If the collision history is empty, the method moves to step 532. In step 532, the current fundamental period is rejected and the method does not send back a solution to eliminate resource collisions.

  [0103] If the collision history is not empty, the method proceeds to step 530 where the current collision is rolled back and set to the last removed collision stored in the collision history. The method moves to step 521.

  [0104] In step 521, the counter associated with the current collision is checked. Counter = 2 indicates that the current collision has been resolved twice: forward solution (fast step delay) and reverse solution (slow step delay). The counter indicates that only forward solutions have been attempted. In step 521, if the counter is 2, the method moves to step 526 to remove the current collision from the collision history and roll back the step again. If the counter is 1, the method proceeds to step 522 and attempts to reverse the current collision.

  [0105] Referring back to step 506, if the current collision is in the collision history, the method moves to step 521 to determine whether both forward and reverse solutions have been attempted.

Regular cleaning
[0106] Periodic cleaning is a recipe that is run for every W substrates in one step of the process sequence. Regular cleaning is not performed for each substrate and is therefore not included in the normal schedule process.

  [0107] In one embodiment of the present invention, periodic cleaning is treated as a special case of scheduling. Periodic cleaning is only used to calculate the basic period. The usage period of the bottleneck resource or the basic cycle can be regarded as an elapsed time between the start times of two subsequent substrates. Since the calculated fundamental period is used to supply the substrate to the cluster tool, each chamber (assigned to a single step and being the only chamber of this single step) is calculated per fundamental period To receive the substrate. If the chamber can complete the substrate processing within the calculated fundamental period, there is no chamber collision. Substrate processing in the chamber generally includes recipe time, substrate transfer time, and post processing required by the chamber. In one embodiment, the scheduler includes a periodic cleaning recipe time to calculate the base period, so that the periodic cleaning can also be completed in the chamber before the next substrate arrives.

  [0108] In one embodiment of the present invention, periodic cleaning is not considered when calculating the fundamental period. The scheduler then iterates through all chambers of the system with periodic cleaning requirements to determine whether periodic cleaning can be completed within the calculated basic period in addition to substrate process time, substrate transfer time and any additional cue time. To test. If the periodic cleaning can be completed before the next substrate arrives, the calculated fundamental period may be used for scheduling.

  [0109] If the periodic cleaning cannot be completed before the next substrate arrives, the basic period is increased so that the periodic cleaning can be completed within the increased basic period.

[0110] If there are two or more chambers used in this step, the scheduler considers alternating chamber usage in this step. At a given time, only one chamber is used for all substrates until periodic cleaning is activated. When the periodic cleaning condition is reached, the sequencer begins to deliver the substrate to the next chamber in the group, while the first chamber runs the periodic cleaning recipe. Thus, in a step where N chambers are used, at a given time, (N-1) chambers are performing periodic cleaning while one of the chambers is processing the substrate. This guarantees stall-free execution. However, if the periodic cleaning recipe is so long that it cannot be completed within N times the basic period, the periodic cleaning will stall the system. Stall time
Stall time = (regular cleaning recipe time) −N ^* equal to basic period. The scheduler stalls during the stall time calculated for every N + 1 substrates.

  [0111] FIG. 8 illustrates a flowchart of a method 600 for determining a periodic cleaning schedule according to an embodiment of the present invention.

  [0112] At step 602, for each step requiring periodic chamber cleaning, the method 600 checks whether two or more chambers are being used.

  [0113] If there is only one chamber for the step that requires periodic cleaning, the method will complete the periodic cleaning within the basic period in step 604 in addition to the substrate process time, substrate transfer time and any additional cue time. Test whether it is possible.

  [0114] If the periodic cleaning can be completed within the basic period in addition to the process time, transfer time, and additional queue time, the scheduler executes the periodic cleaning within the basic period, as shown in step 606. Set to

  [0115] However, if the periodic cleaning cannot be completed within the basic period in addition to the process time, transfer time, and additional queue time, the periodic cleaning is performed within the basic period, as shown in step 608. In addition, the scheduler extends the basic period.

  [0116] Referring back to step 602, if there are two or more chambers for a step that requires periodic cleaning, the scheduler moves to step 612 to determine whether the cleaning time is longer than N times the basic period. Where N is the number of chambers available to perform this step.

  [0117] If the cleaning time is shorter than N times the basic period, the periodic cleaning may be arranged using the current basic period.

[0118] If the cleaning time is longer than N times the basic period, the scheduler moves to step 614 and calculates the stall time for each of the N substrates to perform periodic cleaning.
Runtime variation and dynamic schedule adjustment
[0119] In one embodiment of the present invention, a static schedule is created before the sequencer (which controls the process sequence of the cluster tool) is started and used as input to determine the movement of the cluster tool. Prior to initiating any substrate transfer job, the sequencer asks the scheduler to see if further delays are needed to avoid collisions. However, since the actual time it takes to execute a recipe can vary, particularly in endpoint-based recipes, the scheduler also monitors the system while the process sequence is being executed. The scheduler then adjusts the delay calculated in the static schedule based on the actual time. For example, the start time of step k was 100 seconds, and the delay after this step was 30 seconds in the static schedule. If the substrate arrives at the chamber at 102 seconds due to substrate transfer time variation, the scheduler adjusts the sequencer to wait for 28 seconds after the substrate completes the recipe.

Example
[0120] A simple embodiment using the inventive method for scheduling a process sequence is provided. A single cluster tool having three chambers CH1, CH2, CH3 comprises a single blade robot RI for performing all substrate transfers between chambers CH1, CH2, CH3. Two load locks LLA, LLB are used to move the substrate in and out of a single cluster tool. A single blade factory interface robot FI is used to transfer between the cassette and the load locks LLA, LLB.

  [0121] The input sequence and recipe time are shown in Table 7. Table 8 shows the calculation of the relative start and end times and the collision of the movement of the robot R1.

[0122] The relative time in Table 8 is calculated based on the fundamental period calculated in Table 9. As depicted in the “Remarks” column, there are two conflicts that should be resolved to complete the schedule. Solutions are found and are shown in Table 10.

  [0123] Embodiments of the present invention can be embodied as a program product for use with a computer system for controlling a cluster tool configured to execute a process sequence. The program (s) of the program product defines the functions of the embodiments of the present invention and can be included in a variety of signal bearing media. An exemplary signal bearing medium is (i) information permanently stored in a non-writable storage medium (eg, a read-only memory device in a computer such as a CD-ROM disk readable by a CD-ROM drive), (ii ) Changeable information stored in a writable storage medium (eg, a floppy disk in a diskette drive or hard disk drive), or (iii) communicate to the computer via a communication medium such as a computer or telephone network including wireless communication Information, but is not limited to these. The latter embodiment specifically includes information downloadable from the Internet and other networks. Such signal bearing media represent an embodiment of the present invention when carrying computer readable instructions that direct the functionality of the present invention.

  [0124] In general, the routines executed to implement an embodiment of the invention may be part of an operating system, a specific application, a component, a program, a module, an object, or an instruction sequence. More specifically, the routine executed to implement an embodiment of the present invention may be part of an automatic script that is invoked, for example, at initial program load (IPL). The computer program of the present invention typically comprises a plurality of instructions that are converted into a machine-readable format by a native computer, and therefore executable instructions. The program also has variables and data structures that reside locally with respect to the program or are found in memory or storage devices. In addition, the various programs described herein may be identified based on the application embodied in a particular embodiment of the invention. However, the following specific program terms are used for convenience only and, therefore, the invention should be limited to use only in the specific applications identified and / or suggested by such terms. is not.

  [0125] Even though only the cluster tool for generating gate polysilicon is described in this application, the present invention is adaptable to other processing tools capable of performing processing step sequences. One skilled in the art can adapt the present invention in an applicable environment.

  [0126] While the above is directed to embodiments of the invention, other and further embodiments of the invention may be devised without departing from the basic scope thereof, which scope is defined by the claims that follow. Determined.

FIG. 1 schematically illustrates a cluster tool for semiconductor processing according to an embodiment of the present invention. FIG. 2 illustrates a flowchart of a process sequence for depositing the gate stack. FIG. 3A schematically illustrates a flowchart of an exemplary process sequence according to one embodiment of the present invention. FIG. 3B schematically illustrates the route of the substrate processed in the process sequence of FIG. 3A in the cluster tool of FIG. FIG. 4 schematically illustrates a recipe diagram for the schedule table of the process sequence of FIG. 3A without queue time. FIG. 5 schematically illustrates a recipe diagram of the updated schedule table of FIG. 4 according to one embodiment of the present invention. FIG. 6 illustrates a flowchart of a scheduling method according to an embodiment of the present invention. FIG. 7 illustrates a flowchart of a resource collision elimination method according to an embodiment of the present invention. FIG. 8 illustrates a flowchart of a method for determining a periodic cleaning schedule according to an embodiment of the present invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 ... Cluster tool, 101 ... Vacuum-tight process platform, 102 ... Factory interface, 103, 104 ... Vacuum substrate transfer chamber, 105, 107 ... Robot, 106A, 106B ... Substrate transfer platform, 108, 110, 112, 114, 116, 118 ... Processing chamber, 120 ... Load lock chamber, 128 ... Front opening unified pod, 128A, 128B ... FOUP, 138 ... Substrate transfer robot, 140 ... Substrate aligner, 200 ... Process sequence, 202, 204, 205, 206, 208, 210, 410, 420, 430, 440, 450, 460, 470, 480, 490, 502, 504, 506, 508, 510, 512, 514, 516, 518, 520, 521, 522 524,526,528,530,532,602,604,606,608,612,614 ... Step, 400 ... scheduling method, 500 ... resource conflict removal method, a method determines the 600 ... regular cleaning schedule

Claims (15)

A method for scheduling a process sequence comprising:
Determining an individual schedule by allocating resources to execute the process sequence, the individual schedule comprising a start time indicating a time at which the individual substrate starts a plurality of process steps in the process sequence; When,
Calculating a fundamental period, wherein the fundamental period is defined as a period between two continuous substrate start times;
Detecting a resource collision of the schedule generated from the individual schedule and the basic period;
Adjusting the individual schedule to eliminate detected resource conflicts.   The method of claim 1, wherein the step of detecting resource collisions and the step of adjusting the individual schedule are repeated until no resource collisions are detected.   The method of claim 1, wherein adjusting the individual schedule comprises adding a queue time to delay a process step associated with the detected resource collision to be removed.   The method of claim 2, wherein the process step being delayed has an earlier start time compared to another process step associated with the resource conflict to be removed.   The method of claim 3, wherein adjusting the individual schedule further comprises suppressing the cue time within a cue time constraint. Determining the fundamental period comprises:
Calculating usage periods of all the resources allocated to execute the process sequence;
Setting the basic period according to the longest usage period of all the resources allocated to execute the process sequence;
The method of claim 1, comprising: Extending the fundamental period if resource conflicts cannot be eliminated by adjusting the individual schedule; and
Detecting a resource collision of the updated schedule generated from the individual schedule and the extended basic period;
Adjusting the individual schedule to remove the resource conflict of the updated schedule;
The method of claim 1, further comprising: A computer readable medium containing a computer program for scheduling a process sequence for performing operations when executed by a process,
Determining an individual schedule by allocating resources to execute the process sequence, the individual schedule comprising a start time indicating a time at which the individual substrate starts a plurality of process steps in the process sequence; When,
Calculating a fundamental period, wherein the fundamental period is defined as a period between two continuous substrate start times;
Detecting a resource collision of the schedule generated from the individual schedule and the basic period;
Adjusting the individual schedule to eliminate detected resource conflicts;
A computer readable medium comprising:   The steps of detecting resource collisions and adjusting the individual schedule are repeated until no resource collisions are detected, and the step of adjusting the individual schedule is associated with the detected resource collision to be removed. 9. The computer readable medium of claim 8, comprising adding a queue time to delay a process step. Determining the fundamental period comprises:
Calculating the duration of use of all the resources allocated to execute the process sequence;
Setting the base period according to the longest usage period of all the resources allocated to execute the process sequence;
The computer-readable medium of claim 8, comprising: Extending the fundamental period if resource conflicts cannot be eliminated by adjusting the individual schedule; and
Detecting a resource collision of the updated schedule generated from the individual schedule and the extended basic period;
Adjusting the individual schedule to remove the resource conflict of the updated schedule;
The computer-readable medium of claim 8, further comprising: A method for scheduling a processing sequence, comprising:
A step of generating a processing schedule, wherein there is no waiting period for each of the plurality of processing steps in the processing sequence;
Determining a basic period according to a bottleneck resource usage period;
Detecting a resource collision of the processing schedule based on the basic period;
Adjusting at least one of the processing schedule and the fundamental period to remove the detected resource conflict;
A method comprising:   The method of claim 12, wherein detecting a resource collision comprises detecting a collision of resources occupied by two or more steps in the processing sequence.   13. The method of claim 12, wherein the adjusting step comprises inserting a cue time to delay a step associated with a resource collision to be removed. The adjustment step
Inserting a cue time into one or more processing steps to remove the detected resource conflict;
Detecting a resource collision of the adjusted processing schedule based on the basic period;
Extending the fundamental period when the detected collision recurs;
The method of claim 12 comprising:

Images